Case Study: iwoca achieves reliable, explainable credit decisioning with Chalk

iwoca cuts training data regeneration from 24 hours to under 60 minutes with Chalk

iwoca, a leading European SME lender, faced significant challenges with its data infrastructure for credit underwriting. Their feature pipelines were fragmented, with separate logic for training and production, leading to data drift and unreliable overnight ETL jobs. This eroded trust in their data, and even small timestamp errors could impact high-value credit decisions. They chose Chalk for its feature store to solve these issues of temporal precision and consistency.

Chalk implemented its compute-first architecture, which uses a unified resolver graph to ensure consistent feature definitions across training and production. This allowed iwoca to recompute features directly from source-of-truth data for any point in time, eliminating drift. The results were substantial: training data regeneration time dropped from 24 hours to under 60 minutes. Chalk's solution provided iwoca with guaranteed data integrity, turning accuracy and explainability into system-level guarantees for their critical lending decisions.

iwoca